Runtime Contract Checking

PythonTA supports runtime checking of type annotations, function preconditions, function postconditions, and class representation invariants. In contrast to the checkers from Pylint and our own custom checkers, the checks described in this section execute when your code is run. This makes it easy to incorporate these checks when executing, debugging, and testing your code.

Quick demo

Here is a minimal example of using the python_ta.contracts module to check a function’s preconditions.

# demo.py
def divide(x: int, y: int) -> int:
    """Return x // y.

    Preconditions:
        - y != 0
    """
    return x // y


if __name__ == '__main__':
    from python_ta.contracts import check_all_contracts
    check_all_contracts()

    divide(10, 0)  # Preconditions on y violated

If we run this file, an AssertionError is raised when we call divide(10, 0):

$ python demo.py
Traceback (most recent call last):
  ...
AssertionError: divide precondition "y != 0" violated for arguments {'x': 10, 'y': 0}.

Similarly, violating a parameter’s type annotation also raises an error:

>>> from demo import divide
>>> divide(10, '2')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  ...
AssertionError: divide argument '2' did not match type annotation for parameter "y: <class 'int'>"

API

The python_ta.contracts module offers two functions for enabling contract checking. The first, check_all_contracts, enables contract checking for all functions and classes defined within a module or set of modules. The second, check_contracts, is a decorator allowing more fine-grained control over which functions/classes have contract checking enabled.

python_ta.contracts.check_all_contracts(*mod_names: str, decorate_main: bool = True) → None

Automatically check contracts for all functions and classes in the given modules.

By default (when called with no arguments), the current module is used.

Parameters:

  • *mod_names – The names of modules to check contracts for. These modules must have been previously imported.

  • decorate_main – True if the module being run (where __name__ == ‘__main__’) should have contracts checked.

python_ta.contracts.check_contracts(func: function, module_names: set[str] | None = None) → function
python_ta.contracts.check_contracts(func: Class, module_names: set[str] | None = None) → Class

A decorator to enable contract checking for a function or class.

When used with a class, all methods defined within the class have contract checking enabled. If module_names is not None, only functions or classes defined in a module whose name is in module_names are checked.

Example

>>> from python_ta.contracts import check_contracts
>>> @check_contracts
... def divide(x: int, y: int) -> int:
...     """Return x // y.
...
...     Preconditions:
...        - y != 0
...     """
...     return x // y

You can pass an object into the function validate_invariants to manually check the representation invariants of the object.

python_ta.contracts.validate_invariants(object)

Check that the representation invariants of obj are satisfied.

You can set the ENABLE_CONTRACT_CHECKING constant to True to enable all contract checking.

python_ta.contracts.ENABLE_CONTRACT_CHECKING = True

Set to True to enable contract checking.

You can set the STRICT_NUMERIC_TYPES constant to False to allow more specific numeric types to be accepted by more general type annotations, as described in PEP 484. For example, this allows int values to be accepted by float type annotations.

python_ta.contracts.STRICT_NUMERIC_TYPES = True

Set to False to allow more specific numeric types to be accepted by more general type annotations.

You can set the DEBUG_CONTRACTS constant to True to enable debugging information to be printed when checking contracts.

python_ta.contracts.DEBUG_CONTRACTS = False

Set to True to display debugging messages when checking contracts.

The following constant is used to make contract checking compatible with PyCharm’s “Run File in Python Console” action.

python_ta.contracts.RENAME_MAIN_TO_PYDEV_UMD = True

Set to False to disable workaround for PyCharm’s “Run File in Python Console” action. In most cases you should not need to change this!

Command Line Interface

The python_ta.contracts CLI can execute a file as __main__ with contracts enabled.

The utmost basic usage of this command is with python -m python_ta.contracts FILE where FILE is the executed Python script such as demo.py. See python -m python_ta.contracts --help for the full list of arguments and options. (Note that you may have to write python3 -m depending on your installation)

Note

The python_ta.contracts CLI command will search the main script for where to begin checking contracts from. This search will only find if blocks that are one line and written like if __name__ == '__main__':, while also accepting varying quotations and whitespaces. Any scripts with an unsupported variation of this if condition will not be recognized nor run.

Specifying contracts

This sections describes the different kinds of contract specifications that PythonTA currently supports.

Functions: type annotations

PythonTA will verify all function parameter and return type annotations specified following the standard in PEP484:

def f(x0: type0, x1: type1 ...) -> return_type:

Whenever the function is called, parameter types are checked before executing the function body, and the return type is checked immediately before the function returns. PythonTA uses the typeguard library to check types.

Functions: custom preconditions

You can write arbitrary preconditions as Python expressions in the function docstring, using the following syntax:

Preconditions:
    - <expr>
    - <expr>
    - <expr>

Each expression is evaluated in the same scope as the function body, in top-down order. If any of the expressions evaluate to a falsey value, an AssertionError is raised. A precondition expression can be followed by a comment using #.

PythonTA ignores expressions that cannot be parsed as valid Python code, or that raise an error when they are evaluated. When DEBUG_CONTRACTS is True, PythonTA will print a warning message when it encounters either of these situations.

Examples:

from python_ta.contracts import check_contracts


@check_contracts
def divide(x: int, y: int) -> int:
    """Return x // y.

    Preconditions:
        - y != 0
    """
    return x // y


@check_contracts
def divide_each(lst1: list[int], lst2: list[int]) -> list[int]:
    """Return the quotient of each number in lst1 divided by the corresponding number in lst2.

    Preconditions:
        - len(lst1) == len(lst2)
        - all(n != 0 for n in lst2)
    """
    return [x // y for x, y in zip(lst1, lst2)]

Note

Type annotations are evaluated before these precondition expressions, and so when you are writing custom preconditions you can assume that the parameters have the correct type.

Functions: custom postconditions

Postconditions for a function can be specified much like preconditions in the function docstring:

Postconditions:
    - <expr>
    - <expr>
    - <expr>

One interesting thing to note about postcondition expressions is that they can include the identifier $return_value, which is used to refer to the return value of the function. Thus, each <expr> in the snippet above can refer to the return value of the function through $return_value, in addition to any other variables and identifiers that are in scope for the function.

Aside from the usage and availability of the $return_value identifier, function postconditions are handled the same way as function preconditions.

Examples:

from python_ta.contracts import check_contracts


@check_contracts
def non_negative_sum(x: int, y: int) -> int:
    """Return x + y. If x + y < 0, then return 0 instead.

    Postconditions:
        - $return_value >= 0
    """
    return max(0, x + y)


@check_contracts
def non_negative_sum_each(lst1: list[int], lst2: list[int]) -> list[int]:
    """Return a list of non-negative sums when a number in lst1 is added to the corresponding number in lst2.
    If the sum is negative for a pair, then it is taken to be 0 for that pair.

    Preconditions:
        - len(lst1) == len(lst2)
    Postconditions:
        - all(num >= 0 for num in $return_value)
    """
    return [non_negative_sum(x, y) for x, y in zip(lst1, lst2)]

Note

Postcondition expressions are evaluated after the type check on the return type is complete. Thus, you can assume that the return value of the function has the correct type when postcondition expressions are evaluated.

Classes: methods

Because all methods are functions, PythonTA will also check type annotations and preconditions for methods.

Note

While it is possible to use the check_contracts decorator on an individual method in a class, we strongly recommend using the decorator on the entire class. This will enable contract checking for all methods in the class, and the additional checks for instance attributes and representation invariants described below.

Classes: attribute types

Both instance attributes and class attributes can have type annotations specified within a class definition, using the following syntax:

class A:
    attr0: type0
    attr1: type1
    ...

When contract checking is enabled for the class, these attribute types are checked (using typeguard) at the following times:

  1. After every method call.

  2. Whenever an attribute is reassigned outside a method of the class.

Examples:

from python_ta.contracts import check_contracts


@check_contracts
class Person:
    """A class representing a person.
    """
    age: int
    name: str

    def __init__(self, name: str, age: int) -> None:
        """Initialize a new person.
        """
        self.name = name
        self.age = age

    def set_age_incorrectly(self) -> None:
        """Set this person's age to a string value."""
        self.age = str(self.age)


if __name__ == '__main__':
    # Create a new Person; name and age types are checked at end of __init__
    p = Person('David', 100)

    # Types are checked after this method call; AssertionError is raised
    p.set_age_incorrectly()

    # Types are checked when attribute is reassigned; AssertionError is raised
    p.name = 100

Note

Other static analysis tools and IDEs understand instance attribute type annotations specified within the initializer, e.g.

class A:
    def __init__(self) -> None:
        self.a : int = 10

However, PythonTA currently does not support this syntax; all attributes must have their type annotations specified at the top level of the class body.

Classes: custom representation invariants

Similar to function preconditions, you can define representation invariants within a class docstring. Each representation invariant is a Python expression involving one or more attributes, and are defined using the following syntax:

Representation Invariants:
    - <expr>
    - <expr>
    - <expr>

Representation invariants are checked after method calls and when attributes are reassigned, immediately after attribute types are checked. Aside from when they are checked, representation invariants are handled the same way as function preconditions.

Examples:

from python_ta.contracts import check_contracts


@check_contracts
class Person:
    """A class representing a person.

    Representation Invariants:
        - self.name.isalpha()
        - self.age >= 0
    """
    age: int
    name: str

    def __init__(self, name: str, age: int) -> None:
        """Initialize a new person.
        """
        self.name = name
        self.age = age

    def decrease_age(self) -> None:
        """Subtract 100 from this person's age."""
        self.age -= 100


if __name__ == '__main__':
    # Create a new Person; representation invariants are checked at end of __init__
    p = Person('David', 10)

    # Representation invariants are checked after this method call; AssertionError is raised
    p.decrease_age()

    # Representation invariants are checked when attribute is reassigned; AssertionError is raised
    p.name = '123'

Technical notes

This section describes some more technical features of PythonTA’s contract checking.

Scope

All custom preconditions, postconditions, and representations are evaluated in the scope where their enclosing function/class is defined. Here is an example where a function’s preconditions refer to both a helper function defined in the same module, and one that’s been imported from a separate module.

from math import sqrt
from python_ta.contracts import check_contracts


@check_contracts
def f(x: float, y: float) -> float:
    """Return x + y.

    Preconditions:
        - x >= 0
        - sqrt(x) >= 3.5
        - is_small(y)
    """
    return x + y


def is_small(y: float) -> bool:
    """Return whether the absolute value of y is less than 1."""
    return abs(y) < 1


if __name__ == '__main__':
    # This raises an AssertionError: sqrt(x) >= 3.5 is False
    f(1, 0)

    # This also raises an AssertionError: is_small(y) is False
    f(100, 2)

Warning

The above example illustrates a currently limitation of PythonTA: the sqrt function must be imported so that it is in scope for the precondition, even though the function isn’t called in the program code. This causes some code checking tools (including PythonTA) to report an “unused import” error for the function.

Inheritance

A subclass “inherits” the attribute type annotations and representation invariants of its superclass(es). That is, when contract checking is enabled on a class, PythonTA will check not just the attribute type annotations and representation invariants defined within the class, but within all of its superclasses as well.

Examples:

from python_ta.contracts import check_contracts


@check_contracts
class Person:
    """A class representing a person.

    Representation Invariants:
      - self.age > 0
      - self.name.isalpha()
    """
    age: int
    name: str

    def __init__(self, name: str, age: int) -> None:
        """Initialize a new person.
        """
        self.name = name
        self.age = age


@check_contracts
class Student(Person):
    """A class representing a student.

    Representation Invariants:
      - len(self.id) == 8
    """
    id: str

    def __init__(self, name: str, age: int, id_: str) -> None:
        """Initialize a new student.
        """
        super().__init__(name, age)
        self.id = id_


if __name__ == '__main__':
    s = Student('David', 20, '01234567')

    # Each of the following attribute assignments fail, violating one of the type annotations
    # or representation invariants in Student or Person.
    s.name = 10
    s.name = ''

    s.age = '10'
    s.age = -1

    s.id = 10
    s.id = ''

All attribute type annotations are checked before all representation invariants. Attribute type annotations and representation invariants are checked based on the reverse method resolution order of their classes. So in the above example, the order of checks made is:

  1. self.age is an int

  2. self.name is a str

  3. self.id is a str

  4. self.age > 0

  5. self.name.isalpha()

  6. len(self.id) == 8

Partial Initialization

Post-condition checks on class attribute types and representation invariants are disabled while initializing an instance of a class. Due to the nature of the approach to disabling these checks, you may encounter that these checks will not run if you pass an already-created instance of an object through an externally defined function named __init__ whose first parameter is self.

class DataClass:
    """
    Representation Invariants:
    - self.my_int > 10

    """
    my_int: int

    def __init__(self) -> None:
        self.my_int = 30

    def annotation_violation(self) -> None:
        """
        Should throw a type annotation error with an int attribute being given a string
        """
        self.my_int = 0


class UnrelatedClass:
    def __init__(self) -> None:
        DataClass.annotation_violation(self)


if __name__ == '__main__':
    import python_ta.contracts
    python_ta.contracts.check_all_contracts()

    my_instance = DataClass()
    try:
        my_instance.annotation_violation()
    except AssertionError:
        print('Properly raised AssertionError')
    else:
        print('Did not raise AssertionError')

    my_instance2 = DataClass()
    try:
        UnrelatedClass.__init__(my_instance2)
    except AssertionError:
        print('Properly raised AssertionError')
    else:
        print('Did not raise AssertionError')

The above example outputs the following

>>> Properly raised AssertionError
>>> Did not raise AssertionError